1 // Copyright (C) 2016 and later: Unicode, Inc. and others. 2 // License & terms of use: http://www.unicode.org/copyright.html 3 /* 4 ******************************************************************************* 5 * 6 * Copyright (C) 2001-2014, International Business Machines 7 * Corporation and others. All Rights Reserved. 8 * 9 ******************************************************************************* 10 * file name: unormcmp.cpp 11 * encoding: US-ASCII 12 * tab size: 8 (not used) 13 * indentation:4 14 * 15 * created on: 2004sep13 16 * created by: Markus W. Scherer 17 * 18 * unorm_compare() function moved here from unorm.cpp for better modularization. 19 * Depends on both normalization and case folding. 20 * Allows unorm.cpp to not depend on any character properties code. 21 */ 22 23 #include "unicode/utypes.h" 24 25 #if !UCONFIG_NO_NORMALIZATION 26 27 #include "unicode/unorm.h" 28 #include "unicode/ustring.h" 29 #include "cmemory.h" 30 #include "normalizer2impl.h" 31 #include "ucase.h" 32 #include "uprops.h" 33 #include "ustr_imp.h" 34 35 U_NAMESPACE_USE 36 37 /* compare canonically equivalent ------------------------------------------- */ 38 39 /* 40 * Compare two strings for canonical equivalence. 41 * Further options include case-insensitive comparison and 42 * code point order (as opposed to code unit order). 43 * 44 * In this function, canonical equivalence is optional as well. 45 * If canonical equivalence is tested, then both strings must fulfill 46 * the FCD check. 47 * 48 * Semantically, this is equivalent to 49 * strcmp[CodePointOrder](NFD(foldCase(s1)), NFD(foldCase(s2))) 50 * where code point order, NFD and foldCase are all optional. 51 * 52 * String comparisons almost always yield results before processing both strings 53 * completely. 54 * They are generally more efficient working incrementally instead of 55 * performing the sub-processing (strlen, normalization, case-folding) 56 * on the entire strings first. 57 * 58 * It is also unnecessary to not normalize identical characters. 59 * 60 * This function works in principle as follows: 61 * 62 * loop { 63 * get one code unit c1 from s1 (-1 if end of source) 64 * get one code unit c2 from s2 (-1 if end of source) 65 * 66 * if(either string finished) { 67 * return result; 68 * } 69 * if(c1==c2) { 70 * continue; 71 * } 72 * 73 * // c1!=c2 74 * try to decompose/case-fold c1/c2, and continue if one does; 75 * 76 * // still c1!=c2 and neither decomposes/case-folds, return result 77 * return c1-c2; 78 * } 79 * 80 * When a character decomposes, then the pointer for that source changes to 81 * the decomposition, pushing the previous pointer onto a stack. 82 * When the end of the decomposition is reached, then the code unit reader 83 * pops the previous source from the stack. 84 * (Same for case-folding.) 85 * 86 * This is complicated further by operating on variable-width UTF-16. 87 * The top part of the loop works on code units, while lookups for decomposition 88 * and case-folding need code points. 89 * Code points are assembled after the equality/end-of-source part. 90 * The source pointer is only advanced beyond all code units when the code point 91 * actually decomposes/case-folds. 92 * 93 * If we were on a trail surrogate unit when assembling a code point, 94 * and the code point decomposes/case-folds, then the decomposition/folding 95 * result must be compared with the part of the other string that corresponds to 96 * this string's lead surrogate. 97 * Since we only assemble a code point when hitting a trail unit when the 98 * preceding lead units were identical, we back up the other string by one unit 99 * in such a case. 100 * 101 * The optional code point order comparison at the end works with 102 * the same fix-up as the other code point order comparison functions. 103 * See ustring.c and the comment near the end of this function. 104 * 105 * Assumption: A decomposition or case-folding result string never contains 106 * a single surrogate. This is a safe assumption in the Unicode Standard. 107 * Therefore, we do not need to check for surrogate pairs across 108 * decomposition/case-folding boundaries. 109 * 110 * Further assumptions (see verifications tstnorm.cpp): 111 * The API function checks for FCD first, while the core function 112 * first case-folds and then decomposes. This requires that case-folding does not 113 * un-FCD any strings. 114 * 115 * The API function may also NFD the input and turn off decomposition. 116 * This requires that case-folding does not un-NFD strings either. 117 * 118 * TODO If any of the above two assumptions is violated, 119 * then this entire code must be re-thought. 120 * If this happens, then a simple solution is to case-fold both strings up front 121 * and to turn off UNORM_INPUT_IS_FCD. 122 * We already do this when not both strings are in FCD because makeFCD 123 * would be a partial NFD before the case folding, which does not work. 124 * Note that all of this is only a problem when case-folding _and_ 125 * canonical equivalence come together. 126 * (Comments in unorm_compare() are more up to date than this TODO.) 127 */ 128 129 /* stack element for previous-level source/decomposition pointers */ 130 struct CmpEquivLevel { 131 const UChar *start, *s, *limit; 132 }; 133 typedef struct CmpEquivLevel CmpEquivLevel; 134 135 /** 136 * Internal option for unorm_cmpEquivFold() for decomposing. 137 * If not set, just do strcasecmp(). 138 */ 139 #define _COMPARE_EQUIV 0x80000 140 141 /* internal function */ 142 static int32_t 143 unorm_cmpEquivFold(const UChar *s1, int32_t length1, 144 const UChar *s2, int32_t length2, 145 uint32_t options, 146 UErrorCode *pErrorCode) { 147 const Normalizer2Impl *nfcImpl; 148 const UCaseProps *csp; 149 150 /* current-level start/limit - s1/s2 as current */ 151 const UChar *start1, *start2, *limit1, *limit2; 152 153 /* decomposition and case folding variables */ 154 const UChar *p; 155 int32_t length; 156 157 /* stacks of previous-level start/current/limit */ 158 CmpEquivLevel stack1[2], stack2[2]; 159 160 /* buffers for algorithmic decompositions */ 161 UChar decomp1[4], decomp2[4]; 162 163 /* case folding buffers, only use current-level start/limit */ 164 UChar fold1[UCASE_MAX_STRING_LENGTH+1], fold2[UCASE_MAX_STRING_LENGTH+1]; 165 166 /* track which is the current level per string */ 167 int32_t level1, level2; 168 169 /* current code units, and code points for lookups */ 170 UChar32 c1, c2, cp1, cp2; 171 172 /* no argument error checking because this itself is not an API */ 173 174 /* 175 * assume that at least one of the options _COMPARE_EQUIV and U_COMPARE_IGNORE_CASE is set 176 * otherwise this function must behave exactly as uprv_strCompare() 177 * not checking for that here makes testing this function easier 178 */ 179 180 /* normalization/properties data loaded? */ 181 if((options&_COMPARE_EQUIV)!=0) { 182 nfcImpl=Normalizer2Factory::getNFCImpl(*pErrorCode); 183 } else { 184 nfcImpl=NULL; 185 } 186 if((options&U_COMPARE_IGNORE_CASE)!=0) { 187 csp=ucase_getSingleton(); 188 } else { 189 csp=NULL; 190 } 191 if(U_FAILURE(*pErrorCode)) { 192 return 0; 193 } 194 195 /* initialize */ 196 start1=s1; 197 if(length1==-1) { 198 limit1=NULL; 199 } else { 200 limit1=s1+length1; 201 } 202 203 start2=s2; 204 if(length2==-1) { 205 limit2=NULL; 206 } else { 207 limit2=s2+length2; 208 } 209 210 level1=level2=0; 211 c1=c2=-1; 212 213 /* comparison loop */ 214 for(;;) { 215 /* 216 * here a code unit value of -1 means "get another code unit" 217 * below it will mean "this source is finished" 218 */ 219 220 if(c1<0) { 221 /* get next code unit from string 1, post-increment */ 222 for(;;) { 223 if(s1==limit1 || ((c1=*s1)==0 && (limit1==NULL || (options&_STRNCMP_STYLE)))) { 224 if(level1==0) { 225 c1=-1; 226 break; 227 } 228 } else { 229 ++s1; 230 break; 231 } 232 233 /* reached end of level buffer, pop one level */ 234 do { 235 --level1; 236 start1=stack1[level1].start; /*Not uninitialized*/ 237 } while(start1==NULL); 238 s1=stack1[level1].s; /*Not uninitialized*/ 239 limit1=stack1[level1].limit; /*Not uninitialized*/ 240 } 241 } 242 243 if(c2<0) { 244 /* get next code unit from string 2, post-increment */ 245 for(;;) { 246 if(s2==limit2 || ((c2=*s2)==0 && (limit2==NULL || (options&_STRNCMP_STYLE)))) { 247 if(level2==0) { 248 c2=-1; 249 break; 250 } 251 } else { 252 ++s2; 253 break; 254 } 255 256 /* reached end of level buffer, pop one level */ 257 do { 258 --level2; 259 start2=stack2[level2].start; /*Not uninitialized*/ 260 } while(start2==NULL); 261 s2=stack2[level2].s; /*Not uninitialized*/ 262 limit2=stack2[level2].limit; /*Not uninitialized*/ 263 } 264 } 265 266 /* 267 * compare c1 and c2 268 * either variable c1, c2 is -1 only if the corresponding string is finished 269 */ 270 if(c1==c2) { 271 if(c1<0) { 272 return 0; /* c1==c2==-1 indicating end of strings */ 273 } 274 c1=c2=-1; /* make us fetch new code units */ 275 continue; 276 } else if(c1<0) { 277 return -1; /* string 1 ends before string 2 */ 278 } else if(c2<0) { 279 return 1; /* string 2 ends before string 1 */ 280 } 281 /* c1!=c2 && c1>=0 && c2>=0 */ 282 283 /* get complete code points for c1, c2 for lookups if either is a surrogate */ 284 cp1=c1; 285 if(U_IS_SURROGATE(c1)) { 286 UChar c; 287 288 if(U_IS_SURROGATE_LEAD(c1)) { 289 if(s1!=limit1 && U16_IS_TRAIL(c=*s1)) { 290 /* advance ++s1; only below if cp1 decomposes/case-folds */ 291 cp1=U16_GET_SUPPLEMENTARY(c1, c); 292 } 293 } else /* isTrail(c1) */ { 294 if(start1<=(s1-2) && U16_IS_LEAD(c=*(s1-2))) { 295 cp1=U16_GET_SUPPLEMENTARY(c, c1); 296 } 297 } 298 } 299 300 cp2=c2; 301 if(U_IS_SURROGATE(c2)) { 302 UChar c; 303 304 if(U_IS_SURROGATE_LEAD(c2)) { 305 if(s2!=limit2 && U16_IS_TRAIL(c=*s2)) { 306 /* advance ++s2; only below if cp2 decomposes/case-folds */ 307 cp2=U16_GET_SUPPLEMENTARY(c2, c); 308 } 309 } else /* isTrail(c2) */ { 310 if(start2<=(s2-2) && U16_IS_LEAD(c=*(s2-2))) { 311 cp2=U16_GET_SUPPLEMENTARY(c, c2); 312 } 313 } 314 } 315 316 /* 317 * go down one level for each string 318 * continue with the main loop as soon as there is a real change 319 */ 320 321 if( level1==0 && (options&U_COMPARE_IGNORE_CASE) && 322 (length=ucase_toFullFolding(csp, (UChar32)cp1, &p, options))>=0 323 ) { 324 /* cp1 case-folds to the code point "length" or to p[length] */ 325 if(U_IS_SURROGATE(c1)) { 326 if(U_IS_SURROGATE_LEAD(c1)) { 327 /* advance beyond source surrogate pair if it case-folds */ 328 ++s1; 329 } else /* isTrail(c1) */ { 330 /* 331 * we got a supplementary code point when hitting its trail surrogate, 332 * therefore the lead surrogate must have been the same as in the other string; 333 * compare this decomposition with the lead surrogate in the other string 334 * remember that this simulates bulk text replacement: 335 * the decomposition would replace the entire code point 336 */ 337 --s2; 338 c2=*(s2-1); 339 } 340 } 341 342 /* push current level pointers */ 343 stack1[0].start=start1; 344 stack1[0].s=s1; 345 stack1[0].limit=limit1; 346 ++level1; 347 348 /* copy the folding result to fold1[] */ 349 if(length<=UCASE_MAX_STRING_LENGTH) { 350 u_memcpy(fold1, p, length); 351 } else { 352 int32_t i=0; 353 U16_APPEND_UNSAFE(fold1, i, length); 354 length=i; 355 } 356 357 /* set next level pointers to case folding */ 358 start1=s1=fold1; 359 limit1=fold1+length; 360 361 /* get ready to read from decomposition, continue with loop */ 362 c1=-1; 363 continue; 364 } 365 366 if( level2==0 && (options&U_COMPARE_IGNORE_CASE) && 367 (length=ucase_toFullFolding(csp, (UChar32)cp2, &p, options))>=0 368 ) { 369 /* cp2 case-folds to the code point "length" or to p[length] */ 370 if(U_IS_SURROGATE(c2)) { 371 if(U_IS_SURROGATE_LEAD(c2)) { 372 /* advance beyond source surrogate pair if it case-folds */ 373 ++s2; 374 } else /* isTrail(c2) */ { 375 /* 376 * we got a supplementary code point when hitting its trail surrogate, 377 * therefore the lead surrogate must have been the same as in the other string; 378 * compare this decomposition with the lead surrogate in the other string 379 * remember that this simulates bulk text replacement: 380 * the decomposition would replace the entire code point 381 */ 382 --s1; 383 c1=*(s1-1); 384 } 385 } 386 387 /* push current level pointers */ 388 stack2[0].start=start2; 389 stack2[0].s=s2; 390 stack2[0].limit=limit2; 391 ++level2; 392 393 /* copy the folding result to fold2[] */ 394 if(length<=UCASE_MAX_STRING_LENGTH) { 395 u_memcpy(fold2, p, length); 396 } else { 397 int32_t i=0; 398 U16_APPEND_UNSAFE(fold2, i, length); 399 length=i; 400 } 401 402 /* set next level pointers to case folding */ 403 start2=s2=fold2; 404 limit2=fold2+length; 405 406 /* get ready to read from decomposition, continue with loop */ 407 c2=-1; 408 continue; 409 } 410 411 if( level1<2 && (options&_COMPARE_EQUIV) && 412 0!=(p=nfcImpl->getDecomposition((UChar32)cp1, decomp1, length)) 413 ) { 414 /* cp1 decomposes into p[length] */ 415 if(U_IS_SURROGATE(c1)) { 416 if(U_IS_SURROGATE_LEAD(c1)) { 417 /* advance beyond source surrogate pair if it decomposes */ 418 ++s1; 419 } else /* isTrail(c1) */ { 420 /* 421 * we got a supplementary code point when hitting its trail surrogate, 422 * therefore the lead surrogate must have been the same as in the other string; 423 * compare this decomposition with the lead surrogate in the other string 424 * remember that this simulates bulk text replacement: 425 * the decomposition would replace the entire code point 426 */ 427 --s2; 428 c2=*(s2-1); 429 } 430 } 431 432 /* push current level pointers */ 433 stack1[level1].start=start1; 434 stack1[level1].s=s1; 435 stack1[level1].limit=limit1; 436 ++level1; 437 438 /* set empty intermediate level if skipped */ 439 if(level1<2) { 440 stack1[level1++].start=NULL; 441 } 442 443 /* set next level pointers to decomposition */ 444 start1=s1=p; 445 limit1=p+length; 446 447 /* get ready to read from decomposition, continue with loop */ 448 c1=-1; 449 continue; 450 } 451 452 if( level2<2 && (options&_COMPARE_EQUIV) && 453 0!=(p=nfcImpl->getDecomposition((UChar32)cp2, decomp2, length)) 454 ) { 455 /* cp2 decomposes into p[length] */ 456 if(U_IS_SURROGATE(c2)) { 457 if(U_IS_SURROGATE_LEAD(c2)) { 458 /* advance beyond source surrogate pair if it decomposes */ 459 ++s2; 460 } else /* isTrail(c2) */ { 461 /* 462 * we got a supplementary code point when hitting its trail surrogate, 463 * therefore the lead surrogate must have been the same as in the other string; 464 * compare this decomposition with the lead surrogate in the other string 465 * remember that this simulates bulk text replacement: 466 * the decomposition would replace the entire code point 467 */ 468 --s1; 469 c1=*(s1-1); 470 } 471 } 472 473 /* push current level pointers */ 474 stack2[level2].start=start2; 475 stack2[level2].s=s2; 476 stack2[level2].limit=limit2; 477 ++level2; 478 479 /* set empty intermediate level if skipped */ 480 if(level2<2) { 481 stack2[level2++].start=NULL; 482 } 483 484 /* set next level pointers to decomposition */ 485 start2=s2=p; 486 limit2=p+length; 487 488 /* get ready to read from decomposition, continue with loop */ 489 c2=-1; 490 continue; 491 } 492 493 /* 494 * no decomposition/case folding, max level for both sides: 495 * return difference result 496 * 497 * code point order comparison must not just return cp1-cp2 498 * because when single surrogates are present then the surrogate pairs 499 * that formed cp1 and cp2 may be from different string indexes 500 * 501 * example: { d800 d800 dc01 } vs. { d800 dc00 }, compare at second code units 502 * c1=d800 cp1=10001 c2=dc00 cp2=10000 503 * cp1-cp2>0 but c1-c2<0 and in fact in UTF-32 it is { d800 10001 } < { 10000 } 504 * 505 * therefore, use same fix-up as in ustring.c/uprv_strCompare() 506 * except: uprv_strCompare() fetches c=*s while this functions fetches c=*s++ 507 * so we have slightly different pointer/start/limit comparisons here 508 */ 509 510 if(c1>=0xd800 && c2>=0xd800 && (options&U_COMPARE_CODE_POINT_ORDER)) { 511 /* subtract 0x2800 from BMP code points to make them smaller than supplementary ones */ 512 if( 513 (c1<=0xdbff && s1!=limit1 && U16_IS_TRAIL(*s1)) || 514 (U16_IS_TRAIL(c1) && start1!=(s1-1) && U16_IS_LEAD(*(s1-2))) 515 ) { 516 /* part of a surrogate pair, leave >=d800 */ 517 } else { 518 /* BMP code point - may be surrogate code point - make <d800 */ 519 c1-=0x2800; 520 } 521 522 if( 523 (c2<=0xdbff && s2!=limit2 && U16_IS_TRAIL(*s2)) || 524 (U16_IS_TRAIL(c2) && start2!=(s2-1) && U16_IS_LEAD(*(s2-2))) 525 ) { 526 /* part of a surrogate pair, leave >=d800 */ 527 } else { 528 /* BMP code point - may be surrogate code point - make <d800 */ 529 c2-=0x2800; 530 } 531 } 532 533 return c1-c2; 534 } 535 } 536 537 static 538 UBool _normalize(const Normalizer2 *n2, const UChar *s, int32_t length, 539 UnicodeString &normalized, UErrorCode *pErrorCode) { 540 UnicodeString str(length<0, s, length); 541 542 // check if s fulfill the conditions 543 int32_t spanQCYes=n2->spanQuickCheckYes(str, *pErrorCode); 544 if (U_FAILURE(*pErrorCode)) { 545 return FALSE; 546 } 547 /* 548 * ICU 2.4 had a further optimization: 549 * If both strings were not in FCD, then they were both NFD'ed, 550 * and the _COMPARE_EQUIV option was turned off. 551 * It is not entirely clear that this is valid with the current 552 * definition of the canonical caseless match. 553 * Therefore, ICU 2.6 removes that optimization. 554 */ 555 if(spanQCYes<str.length()) { 556 UnicodeString unnormalized=str.tempSubString(spanQCYes); 557 normalized.setTo(FALSE, str.getBuffer(), spanQCYes); 558 n2->normalizeSecondAndAppend(normalized, unnormalized, *pErrorCode); 559 if (U_SUCCESS(*pErrorCode)) { 560 return TRUE; 561 } 562 } 563 return FALSE; 564 } 565 566 U_CAPI int32_t U_EXPORT2 567 unorm_compare(const UChar *s1, int32_t length1, 568 const UChar *s2, int32_t length2, 569 uint32_t options, 570 UErrorCode *pErrorCode) { 571 /* argument checking */ 572 if(U_FAILURE(*pErrorCode)) { 573 return 0; 574 } 575 if(s1==0 || length1<-1 || s2==0 || length2<-1) { 576 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR; 577 return 0; 578 } 579 580 UnicodeString fcd1, fcd2; 581 int32_t normOptions=(int32_t)(options>>UNORM_COMPARE_NORM_OPTIONS_SHIFT); 582 options|=_COMPARE_EQUIV; 583 584 /* 585 * UAX #21 Case Mappings, as fixed for Unicode version 4 586 * (see Jitterbug 2021), defines a canonical caseless match as 587 * 588 * A string X is a canonical caseless match 589 * for a string Y if and only if 590 * NFD(toCasefold(NFD(X))) = NFD(toCasefold(NFD(Y))) 591 * 592 * For better performance, we check for FCD (or let the caller tell us that 593 * both strings are in FCD) for the inner normalization. 594 * BasicNormalizerTest::FindFoldFCDExceptions() makes sure that 595 * case-folding preserves the FCD-ness of a string. 596 * The outer normalization is then only performed by unorm_cmpEquivFold() 597 * when there is a difference. 598 * 599 * Exception: When using the Turkic case-folding option, we do perform 600 * full NFD first. This is because in the Turkic case precomposed characters 601 * with 0049 capital I or 0069 small i fold differently whether they 602 * are first decomposed or not, so an FCD check - a check only for 603 * canonical order - is not sufficient. 604 */ 605 if(!(options&UNORM_INPUT_IS_FCD) || (options&U_FOLD_CASE_EXCLUDE_SPECIAL_I)) { 606 const Normalizer2 *n2; 607 if(options&U_FOLD_CASE_EXCLUDE_SPECIAL_I) { 608 n2=Normalizer2::getNFDInstance(*pErrorCode); 609 } else { 610 n2=Normalizer2Factory::getFCDInstance(*pErrorCode); 611 } 612 if (U_FAILURE(*pErrorCode)) { 613 return 0; 614 } 615 616 if(normOptions&UNORM_UNICODE_3_2) { 617 const UnicodeSet *uni32=uniset_getUnicode32Instance(*pErrorCode); 618 FilteredNormalizer2 fn2(*n2, *uni32); 619 if(_normalize(&fn2, s1, length1, fcd1, pErrorCode)) { 620 s1=fcd1.getBuffer(); 621 length1=fcd1.length(); 622 } 623 if(_normalize(&fn2, s2, length2, fcd2, pErrorCode)) { 624 s2=fcd2.getBuffer(); 625 length2=fcd2.length(); 626 } 627 } else { 628 if(_normalize(n2, s1, length1, fcd1, pErrorCode)) { 629 s1=fcd1.getBuffer(); 630 length1=fcd1.length(); 631 } 632 if(_normalize(n2, s2, length2, fcd2, pErrorCode)) { 633 s2=fcd2.getBuffer(); 634 length2=fcd2.length(); 635 } 636 } 637 } 638 639 if(U_SUCCESS(*pErrorCode)) { 640 return unorm_cmpEquivFold(s1, length1, s2, length2, options, pErrorCode); 641 } else { 642 return 0; 643 } 644 } 645 646 #endif /* #if !UCONFIG_NO_NORMALIZATION */ 647